Search Results (2,619)

Drug-resistant tuberculosis remains a major challenge to tuberculosis control, particularly in settings with limited access to conventional molecular testing. This study developed and preliminarily evaluated RR-TB GoldDx and Hr-TB GoldDx, rapid visual assays for detecting rifampicin- and isoniazid-resistance-associated mutations in Mycobacterium tuberculosis. The assays combine recombinase polymerase amplification with unmodified gold nanoparticles (AuNPs) for amplification confirmation and thiol-probe-modified AuNPs for detecting resistance-associated mutations at rpoB codons 516, 526, and 531, katG codon 315, and position −15 of the fabG1–inhA promoter. Evaluation using DNA from 50 M. tuberculosis isolates yielded positive internal-control results for all samples. RR-TB GoldDx achieved accuracies of 84% and 86% against phenotypic drug-susceptibility testing and DNA sequencing, respectively, with Cohen’s kappa values of 0.68 and 0.72. Hr-TB GoldDx achieved 80% accuracy and a kappa value of 0.60 against phenotypic testing, while the inhA promoter and katG assays achieved sequencing-based accuracies of 84% and 86%, respectively. Results were available within 1 h at an estimated reagent cost of THB 500–550 (approximately USD 15–16) per test without conventional thermocycling, real-time fluorescence detection, or gel electrophoresis. These instrument-minimal assays show potential for rapid drug-resistance screening but require validation in larger, genetically diverse cohorts and direct respiratory specimens. Full article

Background/Objectives: Curative-intent surgery for early-stage upper gastrointestinal (GI) cancer causes substantial nutritional, functional, and psychological morbidity; whether an integrated post-treatment nutrition and rehabilitation pathway improves distress and survival is unclear. Methods: We conducted a retrospective cohort study of 130 patients undergoing curative-intent esophagectomy or gastrectomy for early-stage esophagogastric cancer: 65 received a structured post-treatment nutrition and rehabilitation support program (Group A), and 65 received standard care (Group B). The primary outcome was psychological distress trajectory over 12 months (Distress Thermometer; secondarily HADS); overall survival was secondary. Propensity-score matching assessed covariate balance; a multivariable Cox model estimated the adjusted mortality association. Results: Distress scores were similar at 1 month (7.2 vs. 7.4; p = 0.43) and diverged progressively (group × time interaction p < 0.001; 4.9 vs. 6.1 at 12 months; p < 0.001, Cohen’s d ≈ 0.8; 95% CI for the 12-month between-group difference 0.65–1.75 points); HADS-A and HADS-D followed a concordant trajectory (8.4 vs. 10.1; 7.7 vs. 9.3; both p ≤ 0.003). Baseline covariates were reasonably balanced before matching (all standardized mean differences [SMDs] < 0.16) and met the prespecified <0.10 balance threshold for most covariates after matching. Fourteen deaths occurred (3 vs. 11); log-rank testing showed a significant survival difference (p = 0.021), consistent after matching (p = 0.037), and in the adjusted Cox model, support-program receipt was associated with lower mortality (hazard ratio 0.25, 95% CI 0.07–0.90; p = 0.034), though underpowered (14 vs. ~37 events required) and hypothesis-generating. Conclusions: Structured post-treatment nutrition and rehabilitation support was associated with markedly lower psychological distress, the prespecified primary outcome, with the achieved sample exceeding the a priori requirement for the assumed effect size; a secondary, exploratory survival association requires prospective confirmation. These findings support prospective, multicenter evaluation of integrated psycho-oncology and nutrition-rehabilitation care pathways after upper GI cancer surgery and should not be interpreted as demonstrating a causal treatment effect. Full article

The α⁰-thalassemia Southeast Asian (--^SEA) and Thai (--^THAI) deletions comprise a significant global health burden, with high prevalence in Southeast Asia. Conventional methods like polymerase chain reaction (PCR) are limited in resource-poor settings due to complexity, time, and cost. This assay was developed as a rapid and simple option for point-of-care (POC) applications. Three assays specific for wild-type DNA, --^SEA and --^THAI deletions utilized recombinase polymerase amplification (RPA) for rapid DNA amplification at a constant low temperature of 37 °C, with a run time of 10 min, eliminating the need for a thermal cycler. The amplified products were detected using the highly specific CRISPR-Cas12a system and read visually via fluorescence or combined with a lateral flow assay (LFA) in 30 min. Successful engineering of a synthetic PAM site enabled Cas12a recognition of the --^SEA target despite the absence of a suitable natural PAM. The platform demonstrated high accuracy in 74 blinded clinical samples, achieving a diagnostic accuracy of 98.20%, a sensitivity of 94.59%, and a specificity of 100% across 222 per-target evaluations. Detection from non-invasive samples achieved 80% to 100% accuracy, depending on the extraction method and readout format used. Our platform showed potential to meet several ASSURED criteria for diagnostic tests in resource-limited settings, demonstrating high potential for widespread use in genetic screening for α⁰-thalassemia (--^SEA) and (--^THAI). Full article

The quality of dental care is determined not only by the dentist’s technical competence but also by their ability to communicate effectively, listen actively, and establish a relationship of trust with the patient. Patients’ perceptions of dentists’ soft skills represent an essential component of patient-centered care; however, evidence from Romania remains limited. Aim: To evaluate patients’ perceptions of the importance of dentists’ non-technical (soft) skills in private dental practices and to analyze the influence of socio-demographic characteristics on these perceptions. Materials and Methods: The study included 297 patients attending private dental practices in Iași, Romania, who anonymously completed a questionnaire consisting of 20 items rated on a five-point Likert scale. Statistical analysis was performed using SPSS Statistics version 31.0. The Chi-square test and Fisher’s exact test were applied where appropriate. Results: Communication and interpersonal skills were considered the most important competencies by patients (91.3%), followed by cognitive skills (84.2%) and personal characteristics (80.2%), whereas artistic skills received the lowest level of appreciation (49.9%). Communication and interpersonal skills were the most highly valued domain (91.3%), followed by cognitive skills (84.2%) and personal characteristics (80.2%), whereas artistic skills received the lowest ratings (49.9%). Younger patients, urban residents, individuals with higher educational attainment, and those attending regular dental check-ups tended to assign greater importance to several of the non-technical competency domains assessed. The most pronounced age-related difference concerned digital technologies, with acceptance of telemedicine and digital communication decreasing markedly with increasing age. Conclusions: Communication and interpersonal skills were the non-technical competencies most highly valued by patients. The observed differences according to socio-demographic characteristics support the need for an individualized communication approach in dental practice. Full article

Lyme borreliosis (LB) is a tick-borne disease caused by a diverse and expanding group of spirochetes characterized by complex biology and advanced immune evasion mechanisms. It presents a wide range of clinical symptoms affecting multiple organ systems and can lead to persistent complications. The pathogenesis of LB remains incompletely understood, and diagnosis typically relies on serologic assays to detect antibodies against LB. However, the standard two-tiered testing (STTT) algorithm is limited by cross-reactivity, low sensitivity, and delayed results, hindering timely and accurate diagnosis. Although molecular tests are considered the gold standard, their reliance on centralized laboratories can delay critical treatment decisions. This underscores the urgent need for rapid, reliable diagnostic tools, particularly for use at the point of hospital admission. Point-of-care serological and direct antigen testing can provide actionable information, supporting decentralized healthcare systems in diagnosing complex diseases, such as LB. In this study, we developed two fluorescent polarization immunoassays (FPIAs) using IgM and IgG LB-specific synthetic epitopes/peptides to evaluate their diagnostic potential. These FPIAs showed high sensitivity and specificity in detecting IgM or IgG anti-LB antibodies in patient sera within minutes. The fluorescently labeled synthetic peptides produced significant polarization differences between infected and healthy samples, allowing clear discrimination. The FPIA-LB functions as a one-step assay, eliminating the need for secondary antibodies or complex protocols. This work highlights the FPIA technique as a robust, rapid, and efficient tool for LB diagnosis, offering a promising advance in improving early detection and patient outcomes, which could save lives and reduce long-term health complications. Full article

Periodontitis is a chronic, biofilm-induced inflammatory disease characterized by episodic connective tissue destruction and is epidemiologically associated with cardiometabolic conditions, notably type 2 diabetes (T2D) and cardiovascular disease (CVD). Conventional clinical signs do not reliably distinguish quiescent from actively progressing disease, which delays intervention and interdisciplinary referral. Chair-side point-of-care testing (POCT) of oral fluids offers a pragmatic, non-invasive adjunct for real-time assessment of disease activity, not only in dental but also in medical settings. This narrative review, reported in line with the Scale for the Assessment of Narrative Review Articles (SANRA), examines the clinical rationale and current evidence for oral-rinse POCT targeting active matrix metalloproteinase-8 (aMMP-8) and calprotectin. Oral-rinse aMMP-8 reflects disease-active collagenolysis and shows the more robust single-marker performance for periodontitis, whereas oral-fluid calprotectin, a neutrophil-derived S100A8/A9 protein, behaves as a predominantly local inflammatory signal and lacks validated oral-fluid reference ranges. Combining the two markers may add information beyond either alone when used in tandem as a complementary pair, but the supporting data are heterogeneous, largely exploratory, and hypothesis-generating. Dual-biomarker oral-rinse POCT is therefore best framed as a risk-stratification and referral aid rather than a stand-alone diagnostic, with combined interpretation potentially supported by AI technologies. Standardized pre-analytics, transparent threshold reporting, oral-fluid calprotectin reference values, and prospective multicenter validation are required before broad implementation. Full article

Background/Objectives: Family caregivers of people with schizophrenia experience substantial psychological and practical demands associated with long-term caregiving. This study examined whether participation in a web-based digital psychoeducation program was associated with longitudinal changes in coping mechanisms among family caregivers of people with schizophrenia. Methods: A non-randomized quasi-experimental pretest–posttest control-group study was conducted at a mental health hospital in Indonesia. The final analytic sample comprised 202 family caregivers of people with schizophrenia, with 101 participants in the intervention group and 101 in the control group. The intervention comprised 10 structured web-based psychoeducation modules delivered over 10 weeks. Coping was assessed using the Brief COPE Inventory and summarized into problem-focused, emotion-focused, and avoidance coping domains. Linear mixed-effects models were used as the primary adjusted longitudinal analysis, with study group, time, group-by-time interaction, age group, and marital status included as fixed effects. Wilcoxon signed-rank and Mann–Whitney U tests were retained as complementary unadjusted analyses, while baseline-adjusted linear regression models were used as sensitivity analyses. Results: Significant group-by-time interactions were observed for problem-focused coping and emotion-focused coping. The adjusted between-group difference in change was 0.450 points (95% CI: 0.357–0.544; partial ηp² = 0.312; p < 0.001) for problem-focused coping and 0.263 points (95% CI: 0.137–0.390; partial ηp² = 0.078; p < 0.001) for emotion-focused coping. In contrast, no statistically significant differential change was observed for avoidance coping (adjusted difference in change = 0.074 points, 95% CI: −0.055 to 0.204; partial ηp² = 0.006; p = 0.259). Baseline-adjusted sensitivity analyses produced a consistent overall pattern. Conclusions: Participation in the 10-week web-based digital psychoeducation program was associated with greater longitudinal improvements in problem-focused and emotion-focused coping, but not avoidance coping. Given the non-randomized design and baseline differences between groups, these findings should be interpreted as adjusted longitudinal associations rather than definitive evidence of causal efficacy. The program may therefore represent an accessible complementary approach to supporting caregiver adaptation in long-term schizophrenia care. Full article

Background: Effective discharge education is essential for promoting postoperative recovery following transurethral resection of the prostate (TURP); however, conventional discharge education may be insufficient to adequately prepare patients for self-care after discharge. Virtual reality (VR)-assisted education has emerged as an innovative approach that provides immersive and interactive learning experiences. This study aimed to evaluate the effect of VR-assisted discharge education on readiness for hospital discharge in patients undergoing TURP. Methods: A randomized controlled trial was conducted with 68 patients undergoing TURP at a university hospital. Participants were randomly assigned to either an intervention group (n = 34), which received standardized VR-assisted discharge education before surgery in addition to routine care, or a control group (n = 34), which received routine discharge education alone. Readiness for hospital discharge was assessed using the Readiness for Hospital Discharge Scale–Short Form (RHDS-SF) before the intervention and immediately prior to discharge. Data were analyzed using a two-way mixed-design analysis of variance. Results: Mean baseline scores were comparable between groups. Readiness for hospital discharge increased significantly over time in both groups; however, the improvement was significantly greater in the intervention group. Mixed-design ANOVA demonstrated a significant main effect of time (p < 0.001) and a significant time × group interaction (p < 0.001), indicating that VR-assisted discharge education produced greater improvements in discharge readiness than routine discharge education. The RHDS-SF demonstrated acceptable to good internal consistency across the two measurement points (Cronbach’s α = 0.705 at pre-test and 0.807 at post-test). Conclusions: The structured VR-assisted discharge education intervention resulted in greater improvements in perceived readiness for hospital discharge than routine discharge education among patients undergoing TURP. Structured VR-assisted education may be a useful complementary approach to routine discharge education for supporting patients’ preparedness for postoperative self-care. Further multicenter studies with longer follow-up are needed to determine whether these improvements result in sustained clinical benefits. Full article

Amyotrophic Lateral Sclerosis (ALS) is increasingly recognized as a multisystem disorder. However, SOD1-associated ALS has traditionally been regarded as an exclusively motor phenotype. This systematic review challenges that paradigm by characterizing the cognitive and behavioural profiles of SOD1 mutation-related ALS. Following PRISMA guidelines, a systematic PubMed search identified 17 eligible studies. Across the included studies, encompassing 222 patients with SOD1 variants, three of the analyzed subjects fulfilled the diagnostic criteria for frontotemporal dementia (FTD), and a larger proportion exhibited multidomain cognitive deficits, including executive dysfunction, language impairment, and working memory decline. Behavioural disturbances, such as apathy, emotional lability, and mental rigidity, were also prominent. The historical assumption that SOD1 mutations completely spare cognitive networks likely reflects an under-recognition of subtle manifestations, compounded by the limited sensitivity of earlier screening tools. These findings indicate that cognitive and behavioural involvement can occur in SOD1-ALS, highlighting a highly heterogeneous phenotype. Encompassing varying degrees of severity—from subtle differences in neuropsychological test scores to formal FTD diagnoses—the current evidence points towards possible variant-specific phenotypes rather than a uniform cognitive syndrome. Routine multidomain cognitive screening is crucial in this population. Properly characterizing these non-motor features allows for more accurate disease staging, better monitoring of clinical progression, and the implementation of truly personalized care strategies. Full article

Background: Major spine surgery carries an inherent risk of hemodynamic instability due to prone positioning, significant blood loss, and the strict necessity to maintain adequate spinal cord perfusion. Hemodynamic management mainly relies on a reactive approach, treating hypotension only after it occurs, which increases the risk of postoperative complications such as acute kidney injury and ischemic events. This narrative review evaluates the clinical impact, current evidence, and future perspectives of integrating Artificial Intelligence (AI) and Machine Learning (ML) algorithms into perioperative hemodynamic care. Methods: A comprehensive literature search was conducted through PubMed, EMBASE, and the Cochrane Library, spanning from inception to January 2026. The search strategy employed combinations of Medical Subject Headings terms and keywords related to “Artificial Intelligence,” “Machine Learning,” “Hypotension Prediction Index,” “hemodynamic monitoring,” and “major spine surgery.” Studies were selected based on their relevance to predictive hemodynamic algorithms, goal-directed fluid therapy (GDFT), and automated closed-loop systems within the perioperative setting of complex spinal interventions. Results: Five studies show that AI/ML tools can improve hemodynamic management in spine surgery: an hypotension prediction index (HPI)-guided algorithm reduced intraoperative hypotension during prone spinal fusion; a machine learning model accurately predicted massive blood loss in metastatic spinal disease; an AutoML framework linked intraoperative hypertension to worse neurological recovery after spinal cord injury (SCI); a case report showed HPI-guided goal-directed therapy enabled safe, transfusion-free major spine surgery; and topological network analysis identified a narrow optimal mean arterial pressure (MAP) range for neurological recovery after SCI. Collectively, these preliminary findings suggest a potential role for AI/ML in reducing hemodynamic instability and enabling more individualized perioperative management in spine surgery. Rather than converging on a single verdict, these five studies fall into three distinct evidentiary categories when appraised using a structured model-validation (V1–V4) and clinical-translation (T0–T4) framework applied within each category: a real-time monitoring technology (HPI) with a substantial extra-spinal evidence base but a limited spine-specific replication record; a single, externally validated but clinically unproven preoperative prediction model; and two retrospective, hypothesis-generating discovery frameworks that remain exploratory irrespective of surgical domain. Conclusions: The evidence identified does not support a single, unified statement about “AI/ML in spine surgery.” Instead, it points to three distinct situations that warrant separate research priorities: consolidating spine-specific replication of an otherwise mature monitoring technology (HPI); externally confirming the clinical utility, rather than only the discriminative accuracy, of a single preoperative prediction model; and prospectively testing the retrospectively derived targets generated by discovery-oriented analytic frameworks. Considered together, these findings should inform hypothesis-driven research design rather than a single implementation-readiness judgment. Full article

Background/Objectives: The integration of artificial intelligence (AI) and digital technologies into dental practice is reshaping clinical workflows, administrative processes, and, increasingly, the patient experience and the doctor–patient relationship. While prior research has documented the attitudes of clinicians and practice managers, the perspective of the patient remains comparatively underexplored. This study examined how dental patients perceive AI integration and digital tools across the dental care pathway, together with the associated implications for data security, cost, and the human dimension of care. Methods: A cross-sectional, questionnaire-based study was conducted among 200 dental patients in Bucharest, Romania, and the surrounding region. The instrument assessed perceived difficulty and availability regarding digital technology, current use of digital tools, demographic and educational characteristics (age, gender, practice environment, educational level), and two attitudinal dimensions, namely digital prudence and concern for technological sustainability, across five subdomains of the dental care pathway: scheduling, diagnosis, treatment planning, feedback, and follow-up (dispensarization). Responses were analyzed using non-parametric tests and exploratory principal component analysis with internal-consistency validation. Results: Patients expressed moderate-to-high interest in AI support during the diagnostic (median = 3.3, IQR = 2.7–3.9) and feedback (median = 3.11, IQR = 2.78–3.67) stages and the lowest interest in scheduling (median = 2.7, IQR = 2.0–3.3). A marked level of digital prudence was observed (median = 3.24, IQR = 2.82–3.61), reflecting concerns about data security, automation, and a possible weakening of the clinician–patient bond. Younger and academically educated patients reported lower perceived difficulty, higher availability, and greater current use of digital tools (all p ≤ 0.001); counterintuitively, the same patients scored significantly higher on digital prudence (Spearman’s ρ = −0.260, p < 0.001). Greater familiarity with digital tools was therefore accompanied by a more critical awareness of their informational risks rather than by uncritical acceptance. Conclusions: Dental patients approach AI through a dual lens of openness and informed caution, welcoming efficiency gains in the clinical and continuity-of-care stages while voicing measured concerns about data security, affordability, and the preservation of human contact. To interpret this profile, we propose two conceptual contributions: a mapping of patient needs onto Maslow’s hierarchy in the context of AI-mediated care and the Informational VUCA framework, which characterizes the volatility, uncertainty, complexity, and ambiguity that patients face when navigating AI-generated information. The findings point to a clear practical agenda of transparent communication, robust data governance, and education strategies adapted to patients’ educational and demographic profiles, so that AI-enhanced workflows strengthen rather than erode the doctor–patient relationship. Full article

Antimicrobial resistance (AMR) is a growing global health threat driven in part by inappropriate and unnecessary antibiotic use resulting from diagnostic uncertainty at the point of care. In outpatient and acute-care settings, clinicians are often unable to rapidly distinguish between viral and bacterial infections, leading to empiric antibiotic prescribing that contributes to the emergence and spread of resistant pathogens. This review examines current and emerging rapid diagnostic technologies for differentiating bacterial and viral infections, including molecular assays, rapid antigen tests, biomarker-based diagnostics, host-response platforms, hematologic methods, and artificial intelligence-based decision-support systems. These technologies are evaluated based on diagnostic accuracy, turnaround time, cost, accessibility, and clinical actionability within real-world healthcare settings. Although several emerging and point-of-care (POC) technologies can provide results within approximately 6–15 min, their ability to consistently align with the timing, workflow, and clinical decision-making requirements of frontline outpatient and emergency-care settings remains variable and incompletely established. Future progress in antimicrobial stewardship will depend on developing rapid, clinically actionable diagnostic systems that integrate seamlessly into patient care and reduce unnecessary antibiotic use. Full article

Background/Objectives: Person-centered care (PCC) and the Technological Competency as Caring in Nursing (TCCN) theory are increasingly emphasized in nursing leadership, but no published study has evaluated a PCC/TCCN training program among Mongolian nurse managers. This uncontrolled, single-group pilot pre–post study examined immediate changes in nurse managers’ self-reported PCC and TCCN perceptions following a one-day collective training program. Methods: The program was delivered on 30 October 2025 during the 2nd International Nursing Symposium in Mongolia. Using convenience sampling, hospitals were invited to send nurse managers who were currently in post or senior nurses who had been formally nominated for such a role. Of 50 attendees, 30 met a pre-specified role-eligibility criterion and formed the matched pre–post analytic sample. Participants completed Mongolian versions of the Person-Centered Care Instrument (PCCI) and the Perceived Technological Competency as Caring in Healthcare Providers Instrument (TCCHI) immediately before and after the program. Pre–post differences were assessed with Wilcoxon signed-rank tests, effect sizes for the Wilcoxon signed-rank test (r = z/√N), and Holm–Bonferroni correction across 16 comparisons. Results: Total and all concept-level PCCI and TCCHI scores were higher after the program than before it, all 16 comparisons remained significant after correction, and the effect sizes were moderate to large (r = 0.41–0.61). The median within-person change was +22.0 points (95% CI 7.0 to 35.5) for the total PCCI score and +24.0 points (95% CI 10.5 to 38.0) for the total TCCHI score (Hodges–Lehmann estimates). Conclusions: In this small convenience sample of Mongolian nurse managers and senior nurses formally nominated for managerial roles, self-reported PCCI and TCCHI scores increased immediately after a one-day training program. Because the study used a single-group design without a control group, employed instruments not yet fully validated in Mongolian, and included no follow-up, these findings are preliminary evidence of an association rather than proof of program effectiveness. Larger, controlled studies with follow-up are needed. Trial Registration: TCTR20260513001, registered retrospectively on 13 May 2026, after data collection had been completed. Full article

Background: Acute brain injury, including traumatic brain injury (TBI), stroke, subarachnoid hemorrhage and secondary neurological injury, such as sepsis-associated encephalopathy (SAE) and delirium, is a major cause of morbidity in the intensive care unit (ICU), and few treatments alter its course once it is established. Critical illness, and brain injury in particular, rapidly disrupts the gut microbiome (GM) and the production of its metabolites. This dysbiosis matters most in neurologically injured patients, because microbial metabolites and a leaky intestinal barrier feed neuroinflammation through the gut–brain axis. Feeding timing and fasting affect circadian biology, the daily rhythm of the GM, and host metabolic pathways, such as ketogenesis, insulin signaling and autophagy. This review asks whether time-restricted or fasting-mimicking strategies can preserve these functions and reduce neuroimmune dysregulation after brain injury. Methods: In this narrative review, we searched the literature for randomized trials, crossover pilot studies, mechanistic human research, and guideline statements comparing continuous, cyclic, and intermittent enteral strategies, and evaluated translational pathways for GM-targeted feeding interventions. Results: Available studies showed no consistent difference in mortality between continuous and intermittent gastric feeding in ICU adults on mechanical ventilation, although the largest pooled analyses report more diarrhea, more abdominal distension and longer ICU stay with intermittent schedules, most pronounced in ventilated patients. While intermittent/cyclic regimens were not associated with an improvement of patient-centered outcomes, pilot studies demonstrated that short macronutrient interruptions (e.g., 12 h) reliably induced a metabolic fasting response in patients with a prolonged ICU stay. One randomized controlled trial (RCT) had GM end-points and reported feasible modulation of gut taxa and 58 differentially abundant serum metabolites with sequential (continuous-to-intermittent) feeding. Only five studies enrolled dedicated brain-injured cohorts; none was designed around neurological or gut–brain mechanistic outcomes, and none measured GM or neuroinflammation. Conclusions: Time-restricted, microbiome-protecting approaches are biologically plausible and deserve a phased translational program to test them in clinical trials. Future trials should focus on brain-injured and other neurologically relevant ICU patients, in whom the gut–brain axis is most engaged, and should pair mechanistic endpoints with clinical safety and proper control for confounders. No clinical study has yet tested whether non-continuous feeding alters neuroinflammatory or neurological outcomes after acute brain injury. There is currently no direct evidence supporting the effectiveness of this intervention in this population, and the case made here is mechanistic, not empirical. Full article

Background/Objectives: Traumatic intracranial hemorrhage (tICH) in older adults following low-energy falls (LEF) represents a common yet diagnostically challenging condition in the emergency department (ED), where predicting injury prior to computed tomography (CT) remains difficult. Machine learning (ML) has been proposed to support CT decision-making, but its feasibility using routinely available pre-CT clinical variables in this specific population remains unclear. This study presents a systematic exploratory benchmarking of ML pipeline configurations for pre-CT tICH prediction in a well-defined retrospective cohort of older emergency patients following LEF. Methods: We performed a secondary analysis from a retrospective observational bicentric study from two university hospital EDs, including 2250 patients aged ≥65 years presenting after an LEF and undergoing cranial CT. Clinical data were extracted manually from electronic health records (EHRs). Eighteen pre-CT clinical features retrieved from electronic health records were selected based on routine availability and ≤10% missingness. Overall, 1224 valid ML pipeline configurations, combining nine classification algorithms, six imputation strategies, four class-balancing approaches, and optional hyperparameter tuning, were evaluated using 10-fold stratified cross-validation on a training set. The 20 highest-ranked configurations by cross-validation AUC were then assessed on a previously inspected exploratory hold-out test set (n = 563); training-derived rule-out operating points were evaluable for 17 of these 20, as three tuned SVM configurations lacked stored out-of-fold predictions. Results: tICH prevalence was 7.0% (n = 158). Across the 20 highest-ranked configurations, hold-out AUC ranged from 0.517 to 0.585, with Matthews correlation coefficient near zero and balanced accuracy of approximately 50% throughout, indicating differences in operating point rather than in discriminative ability. Some of these top-ranked pipelines reached higher cross-validation AUC (up to 0.679) but detected no cases at the default 0.5 threshold—an effect of the decision threshold under class imbalance rather than of the models’ rank-order discrimination, which was itself limited (hold-out AUC of 0.517–0.585). Conclusions: Despite comprehensive exploratory benchmarking across 1224 ML pipelines, routinely available pre-CT clinical features did not provide sufficient discriminatory signal to develop a clinically useful tICH prediction model in this cohort of CT-imaged older adults following LEF. These findings indicate that none of the evaluated configurations produced clinically adequate performance; this near-chance result persisted across all pipelines and most plausibly reflects a combination of limited feature signal, low outcome prevalence, and a sample size below the level required for reliable model development at this event rate. Future studies should target substantially larger prospective multicenter cohorts and evaluate additional feature domains, including structured clinical examination findings, point-of-care biomarkers, and imaging features. Full article